The identification and procurement of the most suitable set of satellite images for each sampling unit have been complex tasks and certainly the single most time-consuming phase of the survey.
After the locations of the 117 sampling units were selected statistically, as described under Section 2.1, using the World Reference System 2 (WRS 2) of the Landsat Program as a reference, the work consisted of defining the historical image (or images) corresponding to the path and row of the recent image and identifying the best acquisitions among those available for both recent and historical coverage.
The historical images, acquired during Landsat mission 1, 2 and 3, correspond to the World Reference System 1 (WRS 1). The two reference systems do not correspond: with mission 4 (July 1982, the beginning of WRS 2) the number of paths was reduced by about one-tenth; owing to this difference and in order to reach the minimum common land area required of one million hectares, in several cases two historical images had to be used.
The selection criteria applied in the identification of the best acquisitions were based on the requirements given below, listed by order of importance:
The optimum season corresponds to that in which forest classes (or, more in general, woody vegetation classes) can best be distinguished from other cover classes. The best time for forest/non-forest separation is usually the beginning of the dry season, where such seasonal condition occur. At that time the spectral signature of the tree cover and that of the undergrowth (grass, herbs) present the highest contrast; the grass has already begun to dry while the tree canopy of deciduous formations is still in full leaf. In order to guide the selection, a special search was conducted to define the optimum season of acquisition for each sampling unit and the information gathered was included in a database.
The closeness of calendar day of acquisition between historical and recent images is very important for areas with strong seasonal fluctuations and deciduous formations. For the analysis of changes it is essential that both historical and recent images present the same vegetative phase and, consequently, the same appearance.
The cloud coverage of historical and recent images combined can reduce considerably the usable area for change assessment, as the common area (the area visible on both images) has already been reduced by the discrepancy resulting from the differing reference systems of WRS 1 and WRS 2. In general a maximum cloud cover of 20 percent was applied when alternatives were available but often the selection was made also in consideration of the location of the clouds (when this precious information was available!) and the size of the common area. The objective was to secure at least one million hectares of visible common land.
An assessment of radiometric quality was usually reported in the image catalogues, together with an assessment of cloud coverage. It should be stated, however, that this information was often unreliable or referred only to macroscopic aspects detected on one band only. The presence of haze and band saturation in old acquisitions are normally not reported in the catalogues, which makes the selection of images a sort of gamble as the actual quality is unknown until the image has been received.
In view of the objective of studying the changes for the standard period 1980–1990, preference was given to the images acquired as close as possible to these two particular years. In the case of a lack of 1980 and 1990 acquisitions and considering that changes become clearer over a longer period of time, the only requirement was that the two images be separated by at least six years.
In the case of missing suitable data due to permanent cloud cover or a simple lack of acquisitions, the approach followed was to order the best image available and integrate it with other satellite data (such as SPOT) and/or existing thematic maps. This ultimate solution became necessary for seven sampling units out of 117 and in each case for only one of the two images needed.
The complete list of images finally selected, procured and analysed in the present survey is given in Annex 9. A total of 283 satellite images were studied, of which 116 represented the recent coverage (only one was missing), 148 represented the historical coverage and 19 were additional images used as “second reference” for SUs already complete with recent and historical images.
The procurement of these images was the result of a cooperative network between the FRA 1990 Project and remote sensing data suppliers, receiving stations, data banks and national institutions throughout the world; a complete list of partecipating institutions and individuals is given in Annex 3 “Project Network”.
The methodology of interpretation and analysis has been designed for implementation in the real world and particularly under the conditions of developing countries and not in isolation under “controlled” conditions. This feature is reflected in several aspects of the methodology and above all in the procedure of interpretation of the satellite images, compilation and analysis of results at sampling unit level.
The dual purpose implicit in this approach can be summarised as follows:
In consideration of the above, the FRA 1990 Project has given much importance to and has concentrated great effort on training and the dissemination of monitoring methodology through numerous workshops and training sessions where the methodology has been presented, discussed and applied practically.
Many institutions have been deeply involved in the application of this methodology, mastering its techniques and appreciating its potential at local level.
With the objective of promoting self reliance in forest resource monitoring and guaranteeing reliable interpretation results the project has organized and carried out four regional workshops and several country-level training sessions where forest monitoring methodology and techniques have been presented, implemented and discussed.
|Bangkok||May 1991||17||9 (Asia)||English|
|Nairobi||November 1991||20||10 (Africa)||English|
|Mexico City||March 1993||21||10 (Latin America)||Spanish|
|Yaoundé||March 1994||27||12 (Africa)||French and English|
|Location||Date||Institution [full names under Table (3.3) 1]|
|Mexico City, Mexico||February 1991||SARH|
|Brasilia, Brazil||March 1991||IBAMA|
|Dehra dun, India||June 1991||FSI|
|Kinshasa, Zaire||August 1991||SPIAF|
|Brasilia, Brazil||March 1992||IBAMA|
|Bangkok, Thailand||April 1992||RFD|
|Jakarta, Indonesia||April 1992||NFI|
|Dehra dun, India||April 1992||FSI|
|Hanoi, Viet Nam||June 1993||FIPI|
|Jakarta, Indonesia||June 1993||NFI|
With the exception of the first workshop in Bangkok, which lasted two weeks, all other Regional Workshops had a duration of three weeks. The training sessions lasted between one and two weeks.
Workshop activities were organized according to the following schedule:
|Presentation of methodology:|
|Compilation and analysis|
|Presentation and discussion of national monitoring activities and other issues|
|Phase II||Distribution of multi-date images to each country team Implementation of monitoring methodology|
|Data entry and processing|
|Phase III||Presentation and discussion of sampling unit results by each country team|
|Discussion on methodological aspects:||Classification|
|Compilation and analysis|
|Conclusions and recommendations|
In order to complete the ambitious planned programme of activities, these workshops always resulted in three weeks of very hard work. There was continuous improvement at each session, especially in the organization of activities and the harmonious balancing of theoretical presentations/discussions with the practical application of the methodology. During the last workshop the activities also included some additional elements of analysis such as the creation of raster maps and an introduction to the standardization of change matrices. This was also possible thanks to the very active participation of the team members from Zaire who were already familiar with the methodology and could thus assist the other participants in all phases of the workshop.
Each workshop concluded with the participants drawing up a set of conclusions and recommendations on the various aspects of the monitoring methodology. These concluding statements, together with a description of workshop's scope, background and agenda, country contributions, activities and results of the sampling units studied were included in four workshop reports. Annex 10 contains some brief presentation sheets, as well as the conclusions and recommendations from each workshop report.
In respect of the three main elements of the methodology, viz, classification, interpretation procedure and compilation system, the conclusions expressed by the participants of the four workshops are rather uniform and can be summarized as follows:
The FRA 1990 classification scheme received a wide range of comments, some calling for the inclusion of more detailed classes and some for a reduction in the number of classes. There seems to be agreement, however, in:
There has been general consensus in considering the interdependent interpretation procedure:
It was also noted, however, that the interdependent interpretation could be simplified if the two images were co-registered spatially and free from relative distortion.
An interesting comment was received from Professor Tape Bidi from the University of Ivory Coast. He affirmed that this procedure of interpretation has a strong educational character and should be introduced in university courses since it stimulates the data interpreter to review critically, evaluate and synthesize several layers of information into a unique logical process.
The compilation system and the type of results produced received positive comments in view of the following main aspects:
The success of the workshops and training session in (always!) completing the analysis of sampling units confirmed that the objectives of adapting to average working conditions were met. Most of the participating foresters/interpreters did not have previous experience in handling computers and software but could master the spreadsheet files and macros easily enough since the assumption that “where there is a Personal Computer there is always someone expert in LOTUS-123®” turned out to be true.
Guidelines, presentation papers and other training material was prepared in three languages (English, French and Spanish) and considerable experience has been acquired in methodology presentation.
Approximately 140 Forestry/Remote Sensing Officers from most tropical countries have been trained in the theoretical and practical aspects of forest resource monitoring. Through these international meetings a network has been established among the participating individuals and institutions on the specific issue of forest monitoring.
The analysis of the sampling units has been carried out, to the largest possible extent, by national satellite data interpreters from the countries where the sampling units were located. The vast majority of these interpreters are foresters by training, experienced interpreters of satellite images and frequently responsible for national vegetation mapping programmes. Annex 2 and Annex 3 provide, respectively, the list of Remote Sensing Lead Centers, where most image interpretation work was carried out, and that of other direct contributors to the analysis of sampling unit data.
Two different approaches were followed for the interpretation and compilation of the 117 sampling units:
Sampling unit analysis was carried out during workshop activities or training sessions after a thorough presentation of methodologies and concepts and under the direct technical supervision of FRA 1990 Project staff.
Groups of sampling units were analysed under contractual and/or cooperation agreements with national institutions by selected teams of interpreters who participated, in most cases, in regional workshops and received additional training at country level.
|SUs analysed at workshops and training sessions||SUs analysed by FRA 1990 Project staff at FAO HQ||Cooperation and Contractual Agreements|
|Institution||Number of SUs|
|LATIN AMERICA||10||1||SARH 4||Mexico||4|
|FIPI 9||Viet Nam||4|
1 The total number of SUs in the above table is 133 and not 117 since several SUs have been analyzed twice.
2 Service Permanent d'Inventaire et d'Aménagement Forestiers (SPIAF), Zaire
3 Istituto Agronomico per l'Oltremare (Overseas Agronomic Institute), Ministry of Foreign Affairs, Italy
4 Secretaria de Agricultura y Recursos Hidraulicos de Mexico (SARH), Mexico
5 Instituto Brasileiro de Meio Ambiente e Recursos Naturais Renováveis (IBAMA), Brazil
6 Javier Anduaga, Ministerio de Agricultura, Oficina General de Planificación Agraria (OPA), Peru
Homero Chaccha Córdova, Instituto Nacional de Recursos Naturales (INRENA), Peru
Leonardo Lugo and Paulino Ruíz Mendoza, Servicio Forestal Venezolano (SEFORVEN), Venezuela
Gerónimo Grimaldez, Centro de Desarrollo Forestal, Ministerio de Asuntos Campesinos y Agropecuarios, Bolivia
Raúl Lara Rico, Centro de Investigaciones de la Capacidad de Uso Mayor de la Tierra (CUMAT), Bolivia
7 Forest Survey of India (FSI), India
8 Royal Forest Department (RFD), Thailand
9 Forest Inventory and Planning Institute (FIPI), Viet Nam
10 National Forest Inventory (NFI), Indonesia
11 Remote Sensing and Resource Data Analysis Dep., National Mapping and Resource Information Authority, Philippines
As is evident from the list of contributors in Table (3.3) 1 the FRA 1990 Project made a major effort to decentralize the work of interpretation and analysis of the sampling units. From the strict point of view of producing results at each sample location, this approach has certainly been uneconomical, given the time needed for coordination, supervision, training and administration. On the other hand, the approach has facilitated the successful dissemination of a standard, consistent and “all weather” methodology to many individuals and institutions and has tapped the most competent sources of field knowledge available.
The role played by Messrs Musampa, Bwangoy and Shoko of SPIAF, Zaire, and Ms Jansen and Mr Lorensi of IBAMA from Brazil, who became so experienced in the methodology that they were able to act as trainers during regional workshop activities, is of particular relevance.
In spite of the two workshops carried out and the positive contribution of SPIAF, the analysis of the African sampling units could not be decentralised totally due to difficulties encountered in travelling to the countries, in establishing links with national institutions and in providing the necessary training. All these “difficult” cases in Africa and, to a lesser extent, in the other regions, have been studied and finalized within the framework of the cooperation activities with the Istituto Agronomico per l'Oltremare (Overseas Agronomic Institute) of Florence, Italy and in particular thanks to the contribution of Ms Angela Dell'Agnello and Ms Ilaria Ambrosini.
The interpretation of satellite data was always carried out with the support of, and with reference to, descriptive reports, including ground photographs, whenever available, and existing cartographic information such as vegetation, forest type, eco-floristic and land use maps; these references were essential when the interpreter was not sufficiently familiar with the field conditions, and proper field verification could not be carried out. This information was mainly used to support decisions on those physiognomic aspects of vegetation that cannot be estimated only on the basis of tone and texture, such as canopy height, to differentiate forest classes from shrubs, or the lower crown densities, to separate open forest from other land cover.
Field verification at sample locations could not be carried out systematically due to the high costs involved and to inaccessibility of a sizeable number of such locations. Field verification has therefore been restricted to a fraction of the sampling units on the basis of (i) importance of ground truth as a function of classification complexity and available knowledge; (ii) accessibility, logistics; and (iii) financial resources.
Table (3.4) 1 reports the status of field knowledge of the present survey by region and ecological zone.
|% of each||3.7||25.9||70.4||18.0||37.7||44.3||17.2||24.1||58.7||14.5||31.6||53.9|
|Levels of field knowledge:|
|Verif.||=||Field verification. Thorough reconnaissance of the study area, with satellite images, from the ground and from light aircraft.|
|Famil.||=||Good familiarity with the exact study area. The interpreter had already visited the study area in the recent past with the purpose of verifying image interpretations.|
|Refer.||=||Knowledge based only on reference material. Knowledge of the field conditions in the study area is based only on knowledge of similar zones and/or maps and reports describing its vegetation and land use|
Table (3.4) 1 shows that field knowledge was not evenly distributed by geographical region and ecological zone. Among the African sampling units, out of 47, only six could be field verified and only seven were familiar to the interpreters.
The situation for Latin America and Asia was much better, where nine and two SUs, respectively, could be field verified and many interpreters were familiar with the sample locations since they are responsible, within their institutions, for the interpretation of satellite data for vegetation mapping. In synthesis, the interpretations were based on good field knowledge (verification or familiarity) in 62.5 percent of the cases for Latin America and 53.3 percent for Asia.
From an ecological perspective, it could be assumed that the interpretation tends to be more difficult in drier than in wetter conditions, as discussed under Section 4.4, making the field knowledge of the dry and moist zones a more valuable ingredient for reliable interpretation than in the wet zone. Following on from this assumption, more efforts have been put into verifying the sampling units of the moist and dry zones, with a resulting “verification intensity” almost five times higher than in the wet zone (18.0 and 17.2 percent versus 3.7). This is also reflected in the share of good field knowledge as the sum of verification and familiarity (last line of the table) where moist (55.7%) and dry (41.3%) well exceed the wet (29.6%), where such field knowledge is less important.
The results of the interpretation and compilation exercises conducted for each SU, produced through cooperation or contractual agreement, have been evaluated carefully and validated at Project Headquarters prior to any further analysis. This evaluation was carried out with the purpose of verifying that:
The first two phases of the evaluation/validation were conducted on the basis of the images and interpretation overlays. The focus of the evaluation was not the delineation of the classes, which is the prerogative of the interpreter, but rather its consistency: interpretation consistency, which means the coherent application of the interpretation key throughout the image (and images), and classification consistency according to the standard classification scheme, which makes the results of all SUs compatible.
The evaluation of the interdependent interpretation of historical and recent images was carried out by overlapping and comparing the two interpretation overlays. These two overlays had to be co-registered spatially, in spite of the relative distortion presented by the images, and the changes had to be clearly evident when comparing one image to the other.
In cases of poor classification or lack of consistency, the SU was re-analysed by the same interpreter after additional training, if this was feasible, or by a new interpreter.
The validation of data entry and compilation was carried out in two phases: (i) using LOTUS123® software to check the files for completeness and proper coding; and (ii) transforming the spreadsheet data into raster maps using IDRISI® in order to visualise the classes and check for data entry errors. The cross-classification of the two state maps (historical and recent), used to produce the change map, allowed one to visualise the location of all changes and compare these changes once more to the two images for a final check.
In the case of wrong or missing codes or errors in data entry, the corrections were made directly at Project Headquarters and then the analysis was finalised.